Overview
- Demis Hassabis told developers at Google I/O that we are “standing in the foothills of the singularity,” a phrase he uses to mean the possible arrival of full artificial general intelligence and which he has said carries about a 50% chance by 2030.
- Google packaged several LLM-driven research systems under a new Gemini for Science brand, including AI Co‑Scientist for hypothesis work and AlphaEvolve for algorithm optimization, and is allowing researchers to apply for access.
- DeepMind pointed to deployed tools such as WeatherNext, which it says gave advance warning for Hurricane Melissa, as examples of real-world scientific impact from AI-driven systems.
- Industry examples and staffing moves show a reorientation toward agentic, generalist systems: independent models have begun making research contributions and senior scientists like John Jumper have shifted from domain‑specific projects to work on coding and agent capabilities.
- Google and DeepMind frame these agentic systems as accelerants or collaborators for scientists, but they note that experimental verification and human oversight remain necessary and that responsibility and access will shape how the shift affects research and public safety.